Our company has been really big on measurement uncertainty lately so this has been one of my big question marks that has come up. If I'm comparing two measurements in a sweep, will I use the VNA uncertainty twice in my RSS? Will it cancel out? If it cancels, what is an acceptable range in a VNA sweep that I can consider this case for? For example a gain/phase variation test where I'm comparing two values X Hz apart, I can't determine how I should handle the VNA uncertainty in these cases.

# Absolute vs. Relative VNA Measurement Uncertainty

Our company has been really big on measurement uncertainty lately so this has been one of my big question marks that has come up. If I'm comparing two measurements in a sweep, will I use the VNA uncertainty twice in my RSS? Will it cancel out? If it cancels, what is an acceptable range in a VNA sweep that I can consider this case for? For example a gain/phase variation test where I'm comparing two values X Hz apart, I can't determine how I should handle the VNA uncertainty in these cases.

To properly compute the uncertainty of two measurements you should take into account the correlation among them. Only in that case it may happen that

some effects cancels out, however to properly account for correlation you should consider all the components affecting each measurement and especially the

correlation effects given by the calibration procedure. The noise is obviously uncorrelated but if you take two measurements with the same calset they are

clearly correlated by the error coefficients.

You have to simplify the matter or use some tool like VNATool and do a lot of manual work to properly take all these effects into account.

If you wish to also include the frequency into the computation ie frequency to frequency correlation then this practically makes the problem too much intesive

and only some experimental work for one port VNA has been don in the literature.

However you can always forget correlation in the RSS method and the answer you will get should overestimate the uncertainty, but it's better than nothing.

Andrea